A Quantitative Analysis of Agentic Pull Requests Across Languages and Task Types

This study presents a quantitative study of AI-generated Pull Requests (Agentic-PRs) to examine acceptance rates, commit structure, and reviewer interactions across 10 programming languages and 12 task types. For analysis, we use the AIDEV_POP dataset, which contains over 456,000 PRs by five AI agents in 61,000 popular GitHub repositories.